Step 1: Understanding the Question:
The question describes an ensemble machine learning technique and asks for its name. The key features described are: 1) it reduces variance, 2) it trains multiple models (trees), and 3) it uses different subsets of data for each model.
Step 2: Detailed Explanation:
Let's analyze the options:
Boosting: This is an ensemble technique where models are trained sequentially. Each subsequent model focuses on correcting the errors made by its predecessor. The primary goal of boosting is to reduce bias, not variance.
Bagging (Bootstrap Aggregating): This technique involves creating multiple random subsets of the original training data with replacement (this is called bootstrapping). A separate model (often a decision tree) is trained independently on each subset. The final prediction is made by aggregating the predictions of all models (e.g., by voting or averaging). This process of averaging over multiple models trained on different data samples is highly effective at reducing the model's variance and preventing overfitting. The Random Forest algorithm is a well-known implementation of Bagging.
Stacking: This is an ensemble technique that combines heterogeneous models by training a "meta-model" to learn how to best combine the predictions from several base models.
Gradient Descent: This is an optimization algorithm used to train a single model by minimizing a loss function. It is not an ensemble technique.
The description in the question perfectly matches the definition of Bagging.
Step 3: Final Answer:
The ensemble technique that reduces variance by training multiple trees on different subsets of data is Bagging.